<MCP.Playground/>
online

Grok Build 0.1

Runs MCP tools

SpaceXAI’s fast coding model, the one behind its Grok Build agent. It is served at over 100 tokens a second, and SpaceXAI says it was trained for agentic coding with MCP support in mind.

Vendor
SpaceXAI
Released
20 May 2026
Context
256K tokens
Input
Text, image, file

Paste a server URL, pick a model, and watch it call your tools in a real conversation. No install.

No server of your own? Leave it blank and use one of the public mock servers.

MCP Evals

Can Grok Build 0.1 actually complete tasks with your server?

One chat shows the tool calls work. An eval writes real tasks from your tool schemas, has Grok Build 0.1 drive each one, and reports where it got the wrong answer even though every call succeeded. Pick Grok Build 0.1 as the driver.

Run evals →See a sample

What Grok Build 0.1 is

Grok Build 0.1 is a coding model SpaceXAI trained for agentic software engineering: web development, debugging and MCP tool use. It powers SpaceXAI’s Grok Build coding agent and went on the API in public beta in May 2026. It has a 256K window and takes text, image and file input. SpaceXAI also pitches it as a fast, cheap option for general tool calling beyond coding. Grok 4.5 has since replaced it as the default model inside Grok Build.

What SpaceXAI says

  • Served at 100+ tokens/s, by SpaceXAI’s figure.
  • Trained specifically for agentic coding tasks, including MCP support.
  • SpaceXAI says it performs best in an agentic harness like Grok Build. It published no benchmark scores with the API release.

These are the vendor’s own claims, not measurements of ours. Run the model against your own server to find out whether they hold for your tools.

How it reaches MCP

  • Grok Build, SpaceXAI’s coding agent, which the model was trained alongside
  • SpaceXAI’s API, which is OpenAI-compatible, so MCP clients that accept a custom endpoint can point at it with a base-URL change
  • MCP Playground’s Agent Studio: paste your server URL above and run Grok Build 0.1 against it in the browser, with no SpaceXAI key

What to watch for

  • SpaceXAI says it performs best in its own harness. Grok Build’s system prompt and tool format are part of what it was trained on, so results in another MCP client can differ from what Grok Build users see.
  • OpenRouter does not list reasoning_effort, parallel_tool_calls or stop for it. You cannot turn its thinking up or down, and multiple calls may come back one per turn.
  • It is version 0.1 and in public beta. SpaceXAI may change the model behind the name, so re-test before comparing with older runs.

Run the same four-step task on Grok Build 0.1 and Grok 4.7 against the complex-schema mock: create_user_profile, process_order, analyze_data, then configure_workflow. Build is the fast one. If it makes the same calls, the speed is free; if not, you can see which step it gets wrong. All the mock servers →

Available here

ModelModel ID
Grok Build 0.1x-ai/grok-build-0.1

Go deeper

Frequently asked questions

What is Grok Build 0.1?
SpaceXAI’s fast coding model, trained for agentic software engineering and MCP tool use. It is the model behind the Grok Build coding agent, and is available on SpaceXAI’s API and OpenRouter in public beta.
Does Grok Build 0.1 support MCP?
Yes. SpaceXAI says it was trained with MCP support, and OpenRouter lists tools and tool_choice. Any client that turns MCP tools into function calls can drive it.
Grok Build 0.1 or Grok 4.7?
Grok Build 0.1 is the fast, cheap coding model. Grok 4.7 is the newer general flagship. Both are available here, so run the same prompt against your server and compare speed and the tool calls.
How fast is Grok Build 0.1?
SpaceXAI serves it at over 100 tokens a second. Speed per tool call also depends on your server’s response time.

Sources

Grok Build 0.1 for MCP: Specs & Tool Calling — Test Free | MCP Playground